Pith. sign in

REVIEW 4 major objections 5 minor 97 references

Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A diagnosis-guided recovery harness lifts agent success and cuts recovery cost.

desk verdict A genuinely useful recipe for turning agent failures into selective recovery, but the paper's own controls undercut the causal claim that diagnosis drives the gains. read the letter →

arxiv 2608.11772 v1 pith:UPFV5VH6 submitted 2026-08-12 cs.CL

Pan Wang , Yihao Hu , Hang Wang , Zirui Lv , Xin Zhang , Jianshe Li , Jiang-Ming Yang , Wei Wu
show 1 more author
Yongqi Tong
This is my paper · ORCID
classification cs.CL
keywords languageagentsself-correctionfailurediagnosisrecoverypolicyactionvalidityretrievalbudgetagentevaluationpromptoptimization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that broad language-agent self-correction fails not because agents lack corrective context but because they lack diagnosis: a coarse task failure mixes incompatible recovery signals, and generic playbooks add irrelevant material that interferes while inflating cost. The authors propose DARC, a harness that profiles development-set failures to identify a task family's dominant failure mode, prunes mismatched interventions from a shared recovery library, and distills a short, verifier-selected success-cost policy that is frozen before deployment. The claim is that this causal order — decide what kind of failure can be repaired, then decide how much recovery evidence to spend — improves average task success over base agents and broad playbooks while reducing environment steps or retrieval budget across ALFWorld, AppWorld, and XBRL Finance. A sympathetic reader would care because the paper offers a practical route to more reliable agents in domains where compiler-like feedback is absent: making failures actionable before making contexts larger.

What carries the argument

The central object is the recovery harness $H_m=(m,\mathcal{R}_m,\pi_m)$, built in two stages. Failure diagnosis profiles development-set failures and prunes the candidate intervention library; policy distillation enumerates bounded intervention chains, scores each by the objective $J(\pi)=\mathrm{dsucc}(\pi)-\lambda\max\{0,\mathrm{dcost}(\pi)-\tau_{\mathrm{free}}\}$, and freezes the winner for deployment. The short-circuit rule invokes a later intervention only after earlier ones fail, with success and cost attributed by the task verifier; the coverage function over interventions is monotone and submodular, and a uniform-convergence bound over the finite chain class explains why pruning the library shrinks search risk.

What would settle it

Run DARC on a benchmark where interventions cannot be evaluated from the same initial state — for example, tasks in which attempting the first intervention irreversibly consumes a shared budget or mutates state before the second intervention starts. If the offline evidence matrix no longer predicts deployment, the distilled short-circuit policy's success should fall toward the generic-playbook level or below the base agent. A complementary test is constructing a task family whose failures split evenly between two distinct modes: the frozen single-mode diagnosis should underperform a full-library cascade on that family.

Watch

Extended reading notes

Core claim

The central discovery is a recovery-harness construction: a frozen tuple $(m, \mathcal{R}_m, \pi_m)$ in which $m$ is a task-family failure diagnosis, $\mathcal{R}_m$ is the admissible intervention subset, and $\pi_m$ is an ordered short-circuit policy selected by training-set verifier feedback to maximize success minus a cost penalty. The paper claims this harness converts ambiguous task failure into typed recovery signals — an action-validity guard for ALFWorld, a procedural fallback chain for AppWorld, a retrieval-budget policy for Finance — and that in each setting it improves average performance (for example, ALFWorld valid_unseen success rises from 39.55% for the base agent and 54.48% for ACE to 90.30%) while reducing environment steps or retrieval budget. It further claims the improvement comes from the diagnosis-restriction pairing itself rather than from privileged information: a matched-information factorial shows the restricted action view and the recovery prompt are each near-inert in isolation and only together produce the large effect, an interaction worth +46.27 percentage points on ALFWorld.

Load-bearing premise

The framework assumes each recovery intervention can be evaluated independently from the same initial task state, with the verifier attributing success and cost to the invoked intervention, so that the offline evidence matrix is a valid proxy for deployment behavior.

Editorial extensions

If this is right

  • On tasks whose failures expose a dominant, offline-profiled failure mode, a frozen diagnosis-guided harness should match or exceed full-library recovery cascades while searching a much smaller policy space; the paper reports this on ALFWorld (99.25% versus 98.51% test accuracy, not significantly different, over a 10x smaller search space).
  • Because the policy is frozen and verifier-selected, it transfers across related splits and tasks when the failure mode is similar: Finance cross-task transfer stays within about 0.5 to 3 percentage points of target-tuned performance, and an ALFWorld action-validity policy distilled on one split reaches 90.30% on the unseen split.
  • Cost-aware ordering means expensive recovery evidence is spent only after cheaper interventions fail: the Finance policy retrieves a mean of 1.5 demonstrations at top macro accuracy, and the ALFWorld harness solves more episodes at every step budget while cutting average environment steps by more than half.
  • The success term is a monotone submodular coverage function, so adding interventions beyond a short policy yields rapidly diminishing recoverable success at growing cost; saturated policy chains are therefore unlikely to add value.
  • The main results degrade gracefully but not uniformly: on AppWorld challenge splits, the ACE baseline remains competitive or better on some metrics, and the paper interprets the gains as average-utility evidence rather than universal dominance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The one-dominant-failure-mode-per-family assumption is likely to break in benchmarks whose episodes mix two failure modes; the natural extension is instance-level routing over multiple diagnoses, and the paper's own frozen design would probably lose to adaptive routing in such settings.
  • The matched-information factorial suggests the mechanism is interface compatibility — the restriction and the prompt only pay off together — so the harness design may transfer to other agent interfaces, such as tool-use or device-control, where the same superadditive interaction could be tested directly.
  • Since policies are selected on training-set verifier feedback, DARC inherits the verifier's quality: under a noisy or partial verifier, or in environments that are not resettable, the offline evidence matrix may systematically mis-estimate deployment cost, and the reported cost savings could shrink.
  • A direct test of the framework's boundary would be a task family where failure modes are balanced rather than dominant; the paper's diagnosis stage predicts a large degradation for the mismatched policy there, whereas a full-library cascade should stay competitive.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes DARC, a diagnosis-guided recovery harness for language agents. The idea is to profile development-set failures to identify a dominant failure mode per task family, restrict the candidate recovery library to interventions that match that failure mode, and then distill a frozen, verifier-scored success-cost fallback policy over the restricted set. The authors evaluate the protocol on ALFWorld (action-validity recovery), AppWorld (procedural-knowledge recovery), and XBRL Finance (format-precision retrieval-budget recovery), reporting consistent average gains over a base agent, ICL, MIPROv2, GEPA, and ACE, together with reduced environment steps or retrieval budget. They also present a matched-information fairness analysis for ALFWorld, a factorial decomposition of the recovery harness, a validation-selected full-library cascade control in Appendix H, and a preliminary weight-space training extension. The paper is transparent about its supervision accounting, its short-circuit policy-estimation assumption, and the scope of its controlled diagnosis isolation experiment.

Significance. If the causal claim held, DARC would be a practical and reproducible recipe for making self-correction selective rather than context-expanding, and the three-domain instantiation would give the claim useful breadth. The paper has genuine strengths: the supervision accounting in Table 2 is explicit; the matched-information analysis in Table 7 directly addresses the strongest fairness objection to the ALFWorld result; the action-view/prompt factorial in Appendix F reports exact task counts with temperature 0; and Appendix H reports a matched full-library cascade control rather than arguing only by ablation. These are the right controls to run. However, the paper's own strongest control — the diagnosis-isolation experiment in Appendix H — shows no statistically significant accuracy benefit from removing diagnosis, and the generic/mismatched ablation in Table 6 is partially degenerate. The significance of the work therefore depends on whether the claims can be reframed as 'diagnosis improves search efficiency and selection stability without sacrificing accuracy' rather than 'diagnosis is what drives the task-success improvements.'

major comments (4)
  1. [Appendix H, Table 14] The only controlled isolation of the diagnosis step does not support the causal claim that pruning mismatched interventions drives held-out task-success gains. The full-library cascade without diagnosis reaches 98.51% on ALFWorld valid_unseen versus 99.25% for the diagnosed cascade, with a 95% CI of [0.00, 2.24] that the paper itself reports as not significant. The measured benefits of diagnosis are a 10x smaller search space and greater selection stability, not higher accuracy. The abstract and Section 1 attribute the headline improvements to making correction selective and pruning mismatched interventions; those statements overstate what Table 14 establishes. The claims should be revised to separate accuracy gains from search-efficiency and stability gains, or additional evidence must show that diagnosis is causally responsible for the task-success improvements.
  2. [Section 6.1, Table 6] The diagnosis-necessity ablation is weakened by the ALFWorld row, where the Generic and Mismatched conditions both report exactly 52.24% success, so the comparison cannot distinguish a generic playbook from a mismatched intervention set on the benchmark with the largest gains. On AppWorld, the Mismatched condition (64.88% TGC) exceeds the Generic condition (61.90% TGC), contrary to the expected ordering stated in the text; the text acknowledges this only after presenting the claim that the correct policy outperforms both. The conclusion that 'correct > generic > mismatched' holds consistently is therefore not supported by the table as presented. The authors should explain the identical ALFWorld values and either provide a non-degenerate generic/mismatched contrast or soften the claim.
  3. [Section 6.2 and Appendix F] The controlled evidence for the diagnosis mechanism is limited to ALFWorld. Appendix F explicitly states that the corresponding factorial decomposition for the AppWorld and Finance harnesses is not available, and Appendix G indicates that the full-library cascade has not been extended to those families. For the two settings where DARC reports its largest average gains, there is no matched validation-selected cascade without diagnosis, so the central claim that diagnosis-guided restriction is responsible for the AppWorld and Finance improvements is unsupported. The generic and mismatched rows in Table 6 for those families do not use the same policy-search protocol as DARC, and their confidence intervals overlap for AppWorld. The authors should either run the full-library cascade on AppWorld and Finance or explicitly restrict the causal claim to ALFWorld.
  4. [Section 4.2, Table 1] The offline evidence-matrix proxy is load-bearing for the distilled policy, and the paper's own explicit assumption is not validated outside resettable environments. The short-circuit estimator in Eqs. (4)-(6) assumes later interventions are evaluated only after an earlier intervention fails and that the verifier can attribute success and cost to the invoked intervention from the same initial state. For ALFWorld this is justified by resettable evaluation, and for Finance by statelessness, but for AppWorld the deployment semantics are described only as 'controlled task attempts' with verifier-attributed success. No experiment demonstrates that the offline matrix predicts deployment outcomes for AppWorld, where intervention attempts may interact with state. The assumption is disclosed, which is good, but a validation check — e.g., comparing offline-selected policy predictions against actual frozen-policy deployment on a held-out set — is needed before the AppWorld distilled policy can be taken as a reliable proxy.
minor comments (5)
  1. [Table 4] The 'Answer Labels' column header is confusing; since it repeats the distinction defined in Table 2, the table should either rename the column to 'Ground-truth answer labels used in adaptation' or add a footnote cross-referencing Table 2.
  2. [Section 5.2, Figure 2] The scenario-cluster bootstrap confidence intervals are reported only for DeepSeek-V4-Flash; the Qwen3.5-27B and Qwen3.6-27B results in Table 3 do not have the same uncertainty quantification, so the statistical-significance claim is not uniform across the three backbones.
  3. [Table 14] The note that the diagnosed row re-selects a cascade over the four action-validity harnesses, and therefore the 99.25% figure is not directly comparable to the fixed single-harness DARC row of 90.30% in Table 3, is easy to miss; this difference in selection protocol should be stated prominently in the main text when Table 14 is discussed.
  4. [Section 6.4, Table 9] The sentence in Section 6.4 says the distilled policy 'stays in the top macro-accuracy band' while reducing the mean budget to 1.5; Table 9 shows that fixed k=2 and k=4 also achieve 94.25%, so the policy-selection advantage over k=2 is only 0.25 pp on macro accuracy, which should be acknowledged in the surrounding text.
  5. [Section 5.3, Table 5] For ALFWorld, the target-tuned row is reported as '–', so the claim that the frozen policy 'preserves much of the target-tuned performance' is not substantiated on that benchmark; the sentence should either report the target-tuned ALFWorld number or restrict the generalization statement to Finance.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DARC's reported gains are held-out evaluations of a training-selected policy, and the self-cited failure-analysis framing is not load-bearing.

full rationale

The paper's derivation chain is self-contained. DARC's recovery policy is selected on training-set verifier feedback (Eqs. 3-7) and then evaluated on held-out splits of ALFWorld, AppWorld, and Finance against external baselines; no reported number is an identity with a fitted parameter by construction. The diagnosis step restricts a pre-specified intervention library, and the policy search is explicitly scored by J(pi) with test labels excluded (Table 2). The only self-citations ([21,76,77]) appear in Related Work as motivation and are not load-bearing. The paper's own strongest control (Appendix H, Table 14) shows diagnosis alone does not significantly change held-out accuracy on ALFWorld (98.51% vs 99.25%, CI [0.00,2.24]), and Appendix F states the factorial decomposition is unavailable for AppWorld and Finance; these are honest limitations on causal attribution of the diagnosis step, not circularity. No equation reduces a claimed prediction to its input, and the paper does not rename a fitted quantity as a prediction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The framework adds no invented entities in the physical sense: all interventions come from a pre-specified library, and the 'recovery harness' is an algorithmic object, not a postulated force, particle, or conserved quantity. The load-bearing free parameters are the undisclosed ALFWorld guard coefficients, the hand-chosen top-12 window, lambda and tau_free in the objective, the chain length L, and the validation-selected Finance retrieval budgets. The key domain assumptions are the single-dominant-failure-mode diagnosis and the short-circuit retry protocol under which the offline evidence matrix is a valid proxy for deployment; both are stated explicitly in Sections 4.2 and 4.3, and both are acknowledged limits of the controlled setting.

free parameters (6)
  • ALFWorld guard score coefficients (Eq. 8) = not reported
    The ranked top-12 action view is the main driver of the ALFWorld gain (ranked recovery 90.30% vs diagnosis-only 38.81%, Table 7), but the six terms S_goal, S_phase, S_search, S_train-prior, S_conflict, S_loop are given without weights or complete definitions; without them the result cannot be reproduced.
  • Action-window size k = 12
    The guard keeps the top 12 admissible actions; the window size is chosen by hand and is load-bearing, since a random top-12 view collapses ACE from 54.48% to 25.37% (Table 7).
  • Cost penalty lambda = 0.02
    Lambda in Eqs. 2 and 7 is set to 0.02 for the matched cascade comparison (Table 14); no sensitivity analysis over lambda is reported.
  • Cost allowance tau_free = not reported
    tau_free in Eqs. 2 and 7 anchors the cost penalty relative to a standard attempt; its value or how it is derived from base-agent cost is not stated.
  • Finance retrieval budgets k = k=2 (Formula), k=1 (Tags), mean budget 1.5
    Retrieval budget is selected from training-set verifier feedback (Section 6.4, Table 9), i.e., fitted on training data. The finding that larger k hurts (k=8, k=16 reduce tag accuracy) is empirical, not predicted in advance.
  • Maximum policy chain length L = 3
    L=3 caps candidate policies in the cascade comparison (Section H, Table 14); a protocol choice that shapes the search space size (40 vs 400 policies).
assumptions (5)
  • domain assumption Each task family instantiates one dominant failure mode that is stable across development and test splits.
    Section 4.1 states 'our experiments evaluate the controlled setting where each family instantiates one dominant failure mode'; Section 4.3 concedes 'transfer across new failure modes remains an empirical question'. If failures mix modes at test time, the frozen diagnosis misfires.
  • domain assumption Interventions can be evaluated under a controlled retry protocol where later interventions start from the same state and the verifier attributes success and cost to the invoked intervention.
    Section 4.2, paragraph after Eq. 7: 'This assumption is explicit because it determines when an offline evidence matrix is a valid proxy for deployment.' The entire policy distillation depends on this proxy being valid.
  • domain assumption Benchmark verifiers (ALFWorld task success, AppWorld TGC/SGC, FiNER/Formula exact-match accuracy) are valid success measures for the claimed task-family failure modes.
    The evidence matrix E_i,j in Eq. 3 inherits the verifiers' definitions. If exact-match scoring is brittle, the Finance 'format precision' diagnosis is an artifact of the metric rather than a property of the task.
  • domain assumption LLM-based failure diagnosis agrees with a verifier-backed reference at 97.22% on a sample of 72 development traces.
    Appendix I validates the diagnosis step on a small sample of 72 traces against a reference that is itself derived from verification signals; the agreement rate is a point estimate with no confidence interval.
  • standard math Coverage functions are monotone and submodular; Hoeffding-union-bound uniform convergence holds at rate O(sqrt(log|Sigma_K|/n)) for the finite policy class.
    Appendix A states these standard results to justify why restricting R_m shrinks the finite-sample search risk. The bound is used only as motivation, not as a performance guarantee.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction." pith.science (2026). https://pith.science/paper/UPFV5VH6

@misc{pith2026260811772,
  author       = {Pith},
  title        = {Pith review of: Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UPFV5VH6}},
  note         = {Machine review of arXiv:2608.11772}
}
read the original abstract

Self-correction is particularly useful when a failure constrains the next repair. Coding agents benefit from this property because compilers, tests, and execution traces turn many failures into typed recovery signals, but broad language-agent tasks often expose only a coarse task failure. This creates a tension for generic recovery playbooks: they broaden the agent's context precisely when the system needs a narrower repair interface, mixing incompatible signals for invalid actions, missing procedures, and strict-format errors. Our insight is that development-set failures can recover part of the missing diagnostic substrate by deciding which recovery interventions are admissible before test-time correction. We propose DARC, a diagnosis-guided recovery harness that profiles task-family failure modes, prunes mismatched interventions from a shared recovery library, and freezes a verifier-selected success-cost policy for deployment. This causal order makes correction selective: the harness first determines what kind of failure can be repaired, then decides how much recovery evidence to spend. In ALFWorld, AppWorld, and XBRL Finance, the same protocol yields an action-validity harness, a procedural-recovery fallback, and a format-precision retrieval policy; in each evaluated setting it improves average task performance over base agents and broad playbooks while reducing environment steps or retrieval budget. Our experiments show that failures need not trigger uniformly more context: DARC turns self-correction from prompt expansion into recovery-interface design. DARC provides a practical route toward more reliable agents in domains where compiler-like feedback is absent: making failures actionable before making contexts larger.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

97 extracted references · 38 canonical work pages

  1. [1]

    Many-shot in-context learning.Advances in Neural Information Processing Systems, 37:76930–76966, 2024

    RishabhAgarwal,AviSingh,LeiZhang,BerndBohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, et al. Many-shot in-context learning.Advances in Neural Information Processing Systems, 37:76930–76966, 2024

    2024

  2. [2]

    Automix: Automatically mixing lan- guagemodels.arXiv preprint arXiv:2310.12963,2023

    Pranjal Aggarwal, Aman Madaan, Ankit Anand, Sriv- idya Pranavi Potharaju, Swaroop Mishra, Pei Zhou, Aditya Gupta, Dheeraj Rajagopal, Karthik Kappagan- thu, Yiming Yang, Shyam Upadhyay, Manaal Faruqui, and Mausam. Automix: Automatically mixing lan- guagemodels.arXiv preprint arXiv:2310.12963,2023

    arXiv

  3. [3]

    Gepa: Reflective prompt evolution can outperform reinforcement learning.arXiv preprint arXiv:2507.19457, 2025

    Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Arnav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, et al. Gepa: Reflective prompt evolution can outperform reinforcement learning.arXiv preprint arXiv:2507.19457, 2025

    arXiv 2025

  4. [4]

    InProceedings of the 62nd Annual Meeting of the Association for Computational Linguis- tics (Volume 1: Long Papers), pages 12248–12267, 2024

    Arash Ahmadian, Chris Cremer, Matthias Gallé, Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ah- metÜstün,andSaraHooker.Backtobasics: Revisiting reinforce-style optimization for learning from human feedback in llms. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguis- tics (Volume 1: Long Papers), pages 12248–12267, 2024

    2024

  5. [5]

    Self-RAG: Learning to retrieve, generate, and critique through self-reflection

    AkariAsai,ZeqiuWu,YizhongWang,AvirupSil,and Hannaneh Hajishirzi. Self-RAG: Learning to retrieve, generate, and critique through self-reflection. InIn- ternational Conference on Learning Representations, 2024. 9 ANT INTERNATIONAL RESEARCH Diagnosis Before Recovery: T urning Agent Failures into Selective Self-Correction

    2024

  6. [6]

    Programsynthesiswithlargelanguagemodels

    JacobAustin,AugustusOdena,MaxwellNye,Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang,CarrieCai,MichaelTerry,QuocLe,andCharles Sutton. Programsynthesiswithlargelanguagemodels. InarXiv preprint arXiv:2108.07732, 2021

    arXiv 2021

  7. [7]

    Digirl: Train- ingin-the-wilddevice-controlagentswithautonomous reinforcement learning.Advances in Neural Informa- tion Processing Systems, 37:12461–12495, 2024

    Hao Bai, Yifei Zhou, Mert Cemri, Jiayi Pan, Alane Suhr, Sergey Levine, and Aviral Kumar. Digirl: Train- ingin-the-wilddevice-controlagentswithautonomous reinforcement learning.Advances in Neural Informa- tion Processing Systems, 37:12461–12495, 2024

    2024

  8. [8]

    Constitutional ai: Harmlessness from ai feedback

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen,AnnaGoldie,AzaliaMirhoseini,CameronMcK- innon, et al. Constitutional ai: Harmlessness from ai feedback. InarXiv preprint arXiv:2212.08073, 2022

    arXiv 2022

Show all 97 references
  1. [9]

    Longbench: A bilingual, multitask benchmark for long context understanding

    Yushi Bai, Xin Lv, Jiajie Zhang, Hongchang Lyu, JiankaiTang,ZhidianHuang,ZhengxiaoDu,XiaoLiu, Aohan Zeng, Lei Hou, et al. Longbench: A bilingual, multitask benchmark for long context understanding. InProceedings of the 62nd annual meeting of the association for computational l...

    2024

  2. [10]

    Codet: Code generation with generated tests.arXiv preprint arXiv:2207.10397, 2022

    Bei Chen, Fengji Zhang, Anh Nguyen, Daoguang Zan, Zeqi Lin, Jian-Guang Lou, and Weizhu Chen. Codet: Code generation with generated tests.arXiv preprint arXiv:2207.10397, 2022

    2022 arXiv

  3. [11]

    Frugal- gpt: How to use large language models while reduc- ing cost and improving performance.arXiv preprint arXiv:2305.05176, 2023

    Lingjiao Chen, Matei Zaharia, and James Zou. Frugal- gpt: How to use large language models while reduc- ing cost and improving performance.arXiv preprint arXiv:2305.05176, 2023

    2023 arXiv

  4. [12]

    Evaluating large language models trained on code

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, HenriquePondedeOliveiraPinto,JaredKaplan,Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brock- man, et al. Evaluating large language models trained on code. InarXiv preprint arXiv:2107.03374, 2021

    2021 arXiv

  5. [13]

    Teaching large language models to self-debug

    Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. Teaching large language models to self-debug. InInternational Conference on Learning Representations, 2024

    2024

  6. [14]

    Training verifiers to solve math word problems

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, MarkChen,HeewooJun,LukaszKaiser,MatthiasPlap- pert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. InarXiv preprint arXiv:2110.14168, 2021

    arXiv

  7. [15]

    Magentic-one: A generalist multi-agent system for solving complex tasks.arXiv preprint arXiv:2411.04468, 2024

    Adam Fourney, Gagan Bansal, Hussein Mozannar, Cheng Tan, Eduardo Salinas, Friederike Niedtner, GraceProebsting,GriffinBassman,JackGerrits,Jacob Alber, et al. Magentic-one: A generalist multi-agent system for solving complex tasks.arXiv preprint arXiv:2411.04468, 2024

    2024 arXiv

  8. [16]

    Precise zero-shot dense retrieval without rel- evance labels

    Luyu Gao, Xueguang Ma, Jimmy Lin, and Jamie Callan. Precise zero-shot dense retrieval without rel- evance labels. InProceedings of the 61st Annual Meeting of the Association for Computational Lin- guistics (Volume 1: Long Papers), pages 1762–1777, 2023

    2023

  9. [17]

    REALM: Retrieval- augmented language model pre-training

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pa- supat, and Ming-Wei Chang. REALM: Retrieval- augmented language model pre-training. InInterna- tional Conference on Machine Learning, pages 3929– 3938, 2020

    2020

  10. [18]

    Measuring massive multitask language under- standing

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Stein- hardt. Measuring massive multitask language under- standing. InInternational Conference on Learning Representations, 2021

    2021

  11. [19]

    MetaGPT: Meta programming for a multi-agent col- laborative framework.International Conference on Learning Representations, 2024

    Sirui Hong, Xiawu Zheng, Jonathan Chen, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, et al. MetaGPT: Meta programming for a multi-agent col- laborative framework.International Conference on Learning Representations, 2024

    2024

  12. [20]

    RULER: What’s the real context size of your long-context language models?arXiv preprint arXiv:2404.06654, 2024

    Cheng-Ping Hsieh, Simeng Sun, Samuel Kriman, Shantanu Acharya, Dima Rekesh, Fei Jia, and Boris Ginsburg. RULER: What’s the real context size of your long-context language models?arXiv preprint arXiv:2404.06654, 2024

    2024 arXiv

  13. [21]

    Seal: Synergistic co-evolution of agents and learning environments.arXiv preprint arXiv:2605.24426, 2026

    Yihao Hu, Zhihao Wen, Xiujin Liu, Pan Wang, Xin Zhang, and Wei Wu. Seal: Synergistic co-evolution of agents and learning environments.arXiv preprint arXiv:2605.24426, 2026

    2026 arXiv

  14. [22]

    Leveraging pas- sage retrieval with generative models for open domain question answering

    Gautier Izacard and Edouard Grave. Leveraging pas- sage retrieval with generative models for open domain question answering. InProceedings of the 16th Con- ference of the European Chapter of the Association for Computational Linguistics, pages 874–880, 2021

    2021

  15. [23]

    Few-shot learning with retrieval augmented language models.arXiv preprint arXiv:2208.03299, 1 (2):4, 2022

    Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi- Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Few-shot learning with retrieval augmented language models.arXiv preprint arXiv:2208.03299, 1 (2):4, 2022

    2022 arXiv

  16. [24]

    Adaptive-rag: Learning to adapt retrieval-augmented large language models through question complexity

    Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, and Jong C Park. Adaptive-rag: Learning to adapt retrieval-augmented large language models through question complexity. InProceedings of the 10 ANT INTERNATIONAL RESEARCH Diagnosis Before Recovery: T urning Agent Failures...

    2024

  17. [25]

    Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan

    Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. SWE-bench: Can language models re- solve real-world github issues?International Confer- ence on Learning Representations, 2024

    2024

  18. [26]

    MRKL systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning

    Eyal Karpas, Omri Abend, Yonatan Belinkov, Barak Lenz,OpherLieber,NirRatner,YoavShoham,Haggai Bata,YoavLevine,KevinLeyton-Brown,etal. MRKL systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning. In...

    2022 arXiv

  19. [27]

    Densepassageretrievalforopen-domain question answering

    VladimirKarpukhin,BarlasOguz,SewonMin,Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tauYih. Densepassageretrievalforopen-domain question answering. InProceedings of the 2020 Con- ference on Empirical Methods in Natural Language Processing, pages 6769–6781, 2020

    2020

  20. [28]

    Dspy: Compiling declarative lan- guagemodelcallsintoself-improvingpipelines.arXiv preprint arXiv:2310.03714, 2023

    Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, ZhiyuanZhang,KeshavSanthanam,SriVardhamanan, SaifulHaq,AshutoshSharma,ThomasTJoshi,Hanna Moazam, et al. Dspy: Compiling declarative lan- guagemodelcallsintoself-improvingpipelines.arXiv preprint arXiv:2310.03714, 2023

    arXiv

  21. [29]

    Decomposedprompting: Amodular approach for solving complex tasks.International Conference on Learning Representations, 2023

    Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabhar- wal. Decomposedprompting: Amodular approach for solving complex tasks.International Conference on Learning Representations, 2023

    2023

  22. [30]

    Language models can solve computer tasks.Advances in Neural Information Processing Systems, 36:39648– 39677, 2023

    Geunwoo Kim, Pierre Baldi, and Stephen McAleer. Language models can solve computer tasks.Advances in Neural Information Processing Systems, 36:39648– 39677, 2023

    2023

  23. [31]

    Large language modelsare zero-shotreasoners

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language modelsare zero-shotreasoners. InAdvances in Neural Information Processing Systems, volume 35, pages 22199–22213, 2022

    2022

  24. [32]

    RLAIF: Scaling re- inforcement learning from human feedback with ai feedback

    Harrison Lee, Sang Michael Xie Phatale, Hassan Mansoor, Thomas Mesnard, Johan Ferret, Kelvin Lu, Colton Bishop, Ethan Hall, Victor Carbune, Abhinav Rastogi, and Sushant Prakash. RLAIF: Scaling re- inforcement learning from human feedback with ai feedback. InarXiv preprint arXi...

    2023 arXiv

  25. [33]

    Retrieval-augmented generation for knowledge- intensive nlp tasks.Advances in neural information processing systems, 33:9459–9474, 2020

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge- intensive nlp tasks.Advances in neural information processing systems...

    2020

  26. [34]

    CAMEL: Communicative agents for mind exploration of large language model society.Advances in Neu- ral Information Processing Systems, 36:51991–52008, 2023

    Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. CAMEL: Communicative agents for mind exploration of large language model society.Advances in Neu- ral Information Processing Systems, 36:51991–52008, 2023

    2023

  27. [35]

    LooGLE: Can long-context language mod- els understand long contexts?arXiv preprint arXiv:2311.04939, 2024

    Jiaqi Li, Mengmeng Wang, Zilong Zheng, and Muhan Zhang. LooGLE: Can long-context language mod- els understand long contexts?arXiv preprint arXiv:2311.04939, 2024

    2024 arXiv

  28. [36]

    API-bank: A comprehensive benchmark for tool-augmented llms

    Minghao Li, Feifan Song, Bowen Yu, Haiyang Yu, Zhoujun Li, Fei Huang, and Yongbin Li. API-bank: A comprehensive benchmark for tool-augmented llms. arXiv preprint arXiv:2304.08244, 2023

    2023 arXiv

  29. [37]

    Rethinking the role of entropy in optimizing tool-use behaviors for large language model agents

    ZepingLi,HongruWang,YiwenZhao,GuanhuaChen, Yixia Li, Keyang Chen, Yixin Cao, Guangnan Ye, Hongfeng Chai, and Zhenfei Yin. Rethinking the role of entropy in optimizing tool-use behaviors for large language model agents. InProceedings of the 64th Annual Meeting of the Associatio...

    2026

  30. [38]

    Remax: A simple, effective, and efficient reinforcement learning method for aligning large language models.arXiv preprint arXiv:2310.10505, 2023

    Ziniu Li, Tian Xu, Yushun Zhang, Zhihang Lin, Yang Yu, Ruoyu Sun, and Zhi-Quan Luo. Remax: A simple, effective, and efficient reinforcement learning method for aligning large language models.arXiv preprint arXiv:2310.10505, 2023

    arXiv

  31. [39]

    Holisticevaluationoflanguagemodels.InTransactions on Machine Learning Research, 2023

    Percy Liang, Rishi Bommasani, Tony Lee, Dimitris Tsipras,DilaraSoylu,MichihiroYasunaga,YianZhang, Deepak Narayanan, Yuhuai Wu, Ananya Kumar, et al. Holisticevaluationoflanguagemodels.InTransactions on Machine Learning Research, 2023

    2023

  32. [40]

    Let’s verify step by step

    Hunter Lightman, Vineet Kosaraju, Yura Burda, Har- rison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let’s verify step by step. InInternational Conference on Learning Representations, 2024

    2024

  33. [41]

    Toxicchat: Unveiling hidden challenges of toxicity detection in real-world user-ai conversation

    Zi Lin, Zihan Wang, Yongqi Tong, Yangkun Wang, Yuxin Guo, Yujia Wang, and Jingbo Shang. Toxicchat: Unveiling hidden challenges of toxicity detection in real-world user-ai conversation. InFindings of the Association for Computational Linguistics: EMNLP 2023, 2023. 11 ANT INTERN...

    2023

  34. [42]

    Liu, Kevin Lin, John Hewitt, Ashwin Paran- jape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paran- jape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. Lost in the middle: How language models use long contexts.Transactions of the Association for Computational Linguistics, 12:157–173, 2024

    2024

  35. [43]

    Agentbench: Evaluating llms as agents

    Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, HanyuLai, YuGu, HangliangDing, KaiwenMen, Kejuan Yang, et al. Agentbench: Evaluating llms as agents. InInternational Conference on Learning Representations, volume 2024, pages 52989–53046, 2024

    2024

  36. [44]

    Agentlite: A lightweight library for building and advancing task-oriented llm agent system.arXiv preprint arXiv:2402.15538, 2024

    Zhiwei Liu, Weiran Yao, Jianguo Zhang, Liangwei Yang, Zuxin Liu, Juntao Tan, Prafulla K Choubey, Tian Lan, Jason Wu, Huan Wang, et al. Agentlite: A lightweight library for building and advancing task-oriented llm agent system.arXiv preprint arXiv:2402.15538, 2024

    2024 arXiv

  37. [45]

    Finer: Finan- cial numeric entity recognition for xbrl tagging

    Lefteris Loukas, Manos Fergadiotis, Ilias Chalkidis, Eirini Spyropoulou, Prodromos Malakasiotis, Ion An- droutsopoulos, and Georgios Paliouras. Finer: Finan- cial numeric entity recognition for xbrl tagging. In Proceedings of the 60th Annual Meeting of the Associ- ation for Co...

    2022

  38. [46]

    Self- refine: Iterative refinement with self-feedback

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan,LuyuGao,SarahWiegreffe,UriAlon,Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al. Self- refine: Iterative refinement with self-feedback. In Advances in Neural Information Processing Systems, volume 36, pages 46534–46594, 2023

    2023

  39. [47]

    Lever: Learning to verify language-to-code genera- tion with execution

    Ansong Ni, Srini Iyer, Dragomir Radev, Veselin Stoy- anov, Wen-tau Yih, Sida Wang, and Xi Victoria Lin. Lever: Learning to verify language-to-code genera- tion with execution. InInternational Conference on Machine Learning, pages 26106–26128. PMLR, 2023

    2023

  40. [48]

    Optimizing instructions and demon- strations for multi-stage language model programs

    Krista Opsahl-Ong, Michael J Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, and Omar Khattab. Optimizing instructions and demon- strations for multi-stage language model programs. InProceedings of the 2024 Conference on Empiri- cal Methods in Natural Langu...

    2024

  41. [49]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. InAdvances in Neural Information Processing Systems, volum...

    2022

  42. [50]

    Understanding and miti- gatingoverrefusalinllmsfromanunveilingperspective of safety decision boundary

    Licheng Pan, Yongqi Tong, Xin Zhang, Xiaolu Zhang, Jun Zhou, and Zhixuan Chu. Understanding and miti- gatingoverrefusalinllmsfromanunveilingperspective of safety decision boundary. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages ...

    2025

  43. [51]

    Optimaltrans- port for llm reward modeling from noisy preference

    Licheng Pan, Haochen Yang, Haoxuan Li, Yunsheng Lu,YongqiTong,YinuoWang,ShijianWang,Zhixuan Chu,LeiShen,YuanLu,andHaoWang. Optimaltrans- port for llm reward modeling from noisy preference. arXiv preprint arXiv:2605.06036, 2026

    2026 arXiv

  44. [52]

    ART: Automatic multi-step rea- soning and tool-use for large language models

    Bhargavi Paranjape, Scott Lundberg, Sameer Singh, Hannaneh Hajishirzi, Luke Zettlemoyer, and Marco Tulio Ribeiro. ART: Automatic multi-step rea- soning and tool-use for large language models. In arXiv preprint arXiv:2303.09014, 2023

    2023 arXiv

  45. [53]

    O’Brien, Carrie J

    Joon Sung Park, Joseph C. O’Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior.arXiv preprint arXiv:2304.03442, 2023

    2023 arXiv

  46. [54]

    Gorilla: Large language model connected with massive apis.Advances in Neural Information Processing Systems, 37:126544–126565, 2024

    Shishir G Patil, Tianjun Zhang, Xin Wang, and Joseph E Gonzalez. Gorilla: Large language model connected with massive apis.Advances in Neural Information Processing Systems, 37:126544–126565, 2024

    2024

  47. [55]

    Webrl: Training llm web agents via self- evolving online curriculum reinforcement learning

    ZehanQi,XiaoLiu,IatLongIong,HanyuLai,Xueqiao Sun, JiadaiSun, XinyueYang, YuYang, ShuntianYao, Wei Xu, et al. Webrl: Training llm web agents via self- evolving online curriculum reinforcement learning. In International Conference on Learning Representations, volume 2025, pages ...

    2025

  48. [56]

    Toolllm: Facilitating large language modelstomaster16000+real-worldapis.InThe twelfth international conference on learning representations, 2023

    Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, et al. Toolllm: Facilitating large language modelstomaster16000+real-worldapis.InThe twelfth international conference on learning representations, 2023

    2023

  49. [57]

    Qwen3.5: Towards native multimodal agents, February 2026

    Qwen Team. Qwen3.5: Towards native multimodal agents, February 2026. URL https://qwen.ai/blog?id= qwen3.5

    2026

  50. [58]

    Qwen3.6-27B: Flagship-level coding in a 27B dense model, April 2026

    Qwen Team. Qwen3.6-27B: Flagship-level coding in a 27B dense model, April 2026. URL https://qwen.ai/ blog?id=qwen3.6-27b

    2026

  51. [59]

    Direct preference optimization: Your language model is secretly a reward model.Advances in neural infor- mation processing systems, 36:53728–53741, 2023

    RafaelRafailov,ArchitSharma,EricMitchell,Christo- pher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model.Advances in neural infor- mation processing systems, 36:53728–53741, 2023. 12 ANT INTERNATIONAL RES...

    2023

  52. [60]

    Tool- former: Language models can teach themselves to use tools.Advances in neural information processing systems, 36:68539–68551, 2023

    TimoSchick,JaneDwivedi-Yu,RobertoDessì,Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettle- moyer, Nicola Cancedda, and Thomas Scialom. Tool- former: Language models can teach themselves to use tools.Advances in neural information processing systems, 36:68539–68551, 2023

    2023

  53. [61]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang,YKLi,YangWu,etal.Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  54. [62]

    HuggingGPT: Solvingaitaskswithchatgptanditsfriendsinhugging face.Advances in Neural Information Processing Systems, 36:38154–38180, 2023

    Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. HuggingGPT: Solvingaitaskswithchatgptanditsfriendsinhugging face.Advances in Neural Information Processing Systems, 36:38154–38180, 2023

    2023

  55. [63]

    Chi, Nathanael Schärli, andDennyZhou

    Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H. Chi, Nathanael Schärli, andDennyZhou. Largelanguagemodelscanbeeasily distracted by irrelevant context. InProceedings of the 40th International Conference on Machine Learning, pages 31210–31227, 2023

    2023

  56. [64]

    Reflexion: Language agents with verbal reinforcement learning

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. Advances in neural information processing systems, 36:8634–8652, 2023

    2023

  57. [65]

    Alfworld: Aligning text and embodied environments for interactive learning.arXiv preprint arXiv:2010.03768, 2020

    Mohit Shridhar, Xingdi Yuan, Marc-Alexandre Côté, Yonatan Bisk, Adam Trischler, and Matthew Hausknecht. Alfworld: Aligning text and embodied environments for interactive learning.arXiv preprint arXiv:2010.03768, 2020

    2010 arXiv

  58. [66]

    Beyond the imitation game: Quantifying and extrapolating the capabilities of lan- guage models.Transactions on Machine Learning Research, 2023

    Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Austin Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of lan- guage models.Transactions on...

    2023

  59. [67]

    Cognitive architectures for language agents.arXiv preprint arXiv:2309.02427, 2023

    TheodoreRSumers,ShunyuYao,KarthikNarasimhan, and Thomas L Griffiths. Cognitive architectures for language agents.arXiv preprint arXiv:2309.02427, 2023

    2023 arXiv

  60. [68]

    Challenging big-bench tasks and whether chain-of-thought can solve them

    Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, AakankshaChowdhery,QuocV.Le,EdH.Chi,Denny Zhou, et al. Challenging big-bench tasks and whether chain-of-thought can solve them. InFindings of the Association for Computational Linguist...

    2023

  61. [69]

    Let me speak freely? a study on the impact of format restrictions on performance of large language models

    Zhi Rui Tam, Cheng-Kuang Wu, Yi-Lin Tsai, Chieh- Yen Lin, Hung-yi Lee, and Yun-Nung Chen. Let me speak freely? a study on the impact of format restrictions on performance of large language models. arXiv preprint arXiv:2408.02442, 2024

    2024 arXiv

  62. [70]

    Eliminating reasoning via inferring with planning: A new frame- worktoguidellms’non-linearthinking.arXiv preprint arXiv:2310.12342, 2023

    Yongqi Tong, Yifan Wang, Dawei Li, Sizhe Wang, Zi Lin, Simeng Han, and Jingbo Shang. Eliminating reasoning via inferring with planning: A new frame- worktoguidellms’non-linearthinking.arXiv preprint arXiv:2310.12342, 2023

    2023 arXiv

  63. [71]

    Canllmslearnfromprevious mistakes? investigating llms’ errors to boost for rea- soning

    Yongqi Tong, Dawei Li, Sizhe Wang, Yujia Wang, Fei Teng, andJingboShang. Canllmslearnfromprevious mistakes? investigating llms’ errors to boost for rea- soning. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p...

    2024

  64. [72]

    Optimizing language model’s rea- soning abilities with weak supervision.arXiv preprint arXiv:2405.04086, 2024

    Yongqi Tong, Sizhe Wang, Dawei Li, Yifan Wang, SimengHan, ZiLin, ChengsongHuang, JiaxinHuang, and Jingbo Shang. Optimizing language model’s rea- soning abilities with weak supervision.arXiv preprint arXiv:2405.04086, 2024

    2024 arXiv

  65. [73]

    Appworld: A controllable world of apps and people for benchmarking interactive coding agents

    Harsh Trivedi, Tushar Khot, Mareike Hartmann, Ruskin Manku, Vinty Dong, Edward Li, Shashank Gupta, Ashish Sabharwal, and Niranjan Balasubrama- nian. Appworld: A controllable world of apps and people for benchmarking interactive coding agents. In Proceedings of the 62nd Annual ...

    2024

  66. [74]

    Finlora: Benchmarking lora methods for fine-tuning llms on financial datasets

    Dannong Wang, Jaisal Patel, Daochen Zha, Steve Y Yang, and Xiao-Yang Liu. Finlora: Benchmarking lora methods for fine-tuning llms on financial datasets. arXiv preprint arXiv:2505.19819, 2025

    2025 arXiv

  67. [75]

    Voyager: Anopen-endedembod- ied agent with large language models.arXiv preprint arXiv:2305.16291, 2023

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Man- dlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and AnimaAnandkumar. Voyager: Anopen-endedembod- ied agent with large language models.arXiv preprint arXiv:2305.16291, 2023

    2023 arXiv

  68. [76]

    REFLEX: Reflective evolution from LLM experience

    Pan Wang. REFLEX: Reflective evolution from LLM experience. InICML 2026 AI for Science Work- shop, 2026. URL https://openreview.net/forum?id= xftknNrW7n

    2026

  69. [77]

    Atlasva: Self-evolving vi- sual skill memory for teacher-free vlm agents.arXiv preprint arXiv:2605.17933, 2026

    Pan Wang, Yihao Hu, Xiujin Liu, Jingchu Yang, Hang Wang, and Zhihao Wen. Atlasva: Self-evolving vi- sual skill memory for teacher-free vlm agents.arXiv preprint arXiv:2605.17933, 2026

    2026 arXiv

  70. [78]

    Renxi Wang, Haonan Li, Xudong Han, Yixuan Zhang, andTimothyBaldwin. Learningfromfailure: Integrat- ingnegativeexampleswhenfine-tuninglargelanguage 13 ANT INTERNATIONAL RESEARCH Diagnosis Before Recovery: T urning Agent Failures into Selective Self-Correction models as agents.a...

    2024 arXiv

  71. [79]

    BPO: Towards balanced preference optimization between knowledge breadth and depth in alignment

    Sizhe Wang, Yongqi Tong, Hengyuan Zhang, Dawei Li, Xin Zhang, and Tianlong Chen. BPO: Towards balanced preference optimization between knowledge breadth and depth in alignment. InProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for C...

    2025

  72. [80]

    Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yue Song, Bowen Li, Jaskirat Singh, et al

    Xingyao Wang, Boxuan Li, Yufan Song, Frank F. Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yue Song, Bowen Li, Jaskirat Singh, et al. OpenDevin: An open platformforaisoftwaredevelopersasgeneralistagents. arXiv preprint arXiv:2407.16741, 2024

    2024 arXiv

  73. [81]

    Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2022

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models.arXiv preprint arXiv:2203.11171, 2022

    2022 arXiv

  74. [82]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, FeiXia, EdChi, QuocV.Le, andDennyZhou. Chain-of-thought prompting elicits reasoning in large language models. InAdvances in Neural Information Processing Systems, volume 35, pages 24824–24837, 2022

    2022

  75. [83]

    Generating sequences by learning to self- correct

    Sean Welleck, Ximing Lu, Peter West, Faeze Brah- man, Tianxiao Shen, Daniel Khashabi, and Yejin Choi. Generating sequences by learning to self- correct. InThe Eleventh International Conference on Learning Representations, 2023. URL https: //openreview.net/forum?id=hH36JeQZDaO

    2023

  76. [84]

    Autogen: Enabling next-genllmapplicationsviamulti-agentconversations

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, et al. Autogen: Enabling next-genllmapplicationsviamulti-agentconversations. InFirst conference on language modeling, 2024

    2024

  77. [85]

    Agentless: Demystifying llm- based software engineering agents.arXiv preprint arXiv:2407.01489, 2024

    Chunqiu Steven Xia, Yinlin Deng, Sam Dunn, and Lingming Zhang. Agentless: Demystifying llm- based software engineering agents.arXiv preprint arXiv:2407.01489, 2024

    2024 arXiv

  78. [86]

    Understanding conflicts in multi- objective alignment through reward consistency

    Zhihao Xu, Yongqi Tong, Xin Zhang, Jun Zhou, and Xiting Wang. Understanding conflicts in multi- objective alignment through reward consistency. In Findings of the Association for Computational Linguis- tics: ACL 2026, pages 5450–5472, 2026

    2026

  79. [87]

    Correctiveretrievalaugmentedgeneration.arXiv preprint arXiv:2401.15884, 2024

    Shi-Qi Yan, Jia-Chen Gu, Yun Zhu, and Zhen-Hua Ling. Correctiveretrievalaugmentedgeneration.arXiv preprint arXiv:2401.15884, 2024

    2024 arXiv

  80. [88]

    SWE-agent: Agent-computer interfaces enable automatedsoftwareengineering.InAdvances in Neural Information Processing Systems, volume 37, 2024

    JohnYang,CarlosEJimenez,AlexanderWettig,Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. SWE-agent: Agent-computer interfaces enable automatedsoftwareengineering.InAdvances in Neural Information Processing Systems, volume 37, 2024

    2024

  81. [89]

    How is llm reasoning distracted by irrelevant context? an analysis using a controlled benchmark

    Minglai Yang, Ethan Huang, Liang Zhang, Mihai Sur- deanu, William Yang Wang, and Liangming Pan. How is llm reasoning distracted by irrelevant context? an analysis using a controlled benchmark. InProceed- ings of the 2025 Conference on Empirical Methods in Natural Language Proc...

    2025

  82. [90]

    React: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629, 2022

    2022 arXiv

  83. [91]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. Tree of thoughts: Deliberate problem solving with large language models. InAdvances in Neural Information Processing Systems, volume 36, pages 11809–11822, 2023

    2023

  84. [92]

    𝜏-bench: : A benchmark for tool-agent- user interaction in real-world domains.arXiv preprint arXiv:2406.12045, 2024

    ShunyuYao, NoahShinn, PedramRazavi, andKarthik Narasimhan. 𝜏-bench: : A benchmark for tool-agent- user interaction in real-world domains.arXiv preprint arXiv:2406.12045, 2024

    2024 arXiv

  85. [93]

    Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah D. Goodman. STaR: Bootstrapping reasoning with rea- soning. InAdvances in Neural Information Processing Systems, volume 35, pages 15476–15488, 2022

    2022

  86. [94]

    Agenticcon- text engineering: Evolving contexts for self-improving language models

    Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton,ChenWu,MengmengJi,HanchenLi,Urmish Thakker,JamesZou,andKunleOlukotun. Agenticcon- text engineering: Evolving contexts for self-improving language models. InThe Fourteenth Int...

    2026

  87. [95]

    Debug like a human: A large language model debugger via verifying runtime execution step-by-step

    Li Zhong, Zilong Wang, and Jingbo Shang. Debug like a human: A large language model debugger via verifying runtime execution step-by-step. InFindings of the Association for Computational Linguistics: ACL 2024, pages 851–870, 2024

    2024

  88. [96]

    Least-to- most prompting enables complex reasoning in large language models

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Olivier Bousquet, Quoc Le, and Ed Chi. Least-to- most prompting enables complex reasoning in large language models. InInternational Conference on Learning Representations, 2023. 14 A...

    2023

  89. [97]

    Archer: Training language model agents via hierarchical multi-turn rl.arXiv preprint arXiv:2402.19446, 2024

    Yifei Zhou, Andrea Zanette, Jiayi Pan, Sergey Levine, and Aviral Kumar. Archer: Training language model agents via hierarchical multi-turn rl.arXiv preprint arXiv:2402.19446, 2024. A Theoretical Proofs A.1 Proof of Monotone Submodularity For an intervention subset𝐴, a task is ...

    2024 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.